A System for Developing Feature Extraction Algorithms for Retinal Implant Devices

Abstract

Purpose: : This work is related to the efforts of the Boston Retinal Implant Project to develop a sub-retinal prosthesis to restore vision to the blind. The specific purpose of this poster is to present a system that may be used to extract features suitable for transmission to a retinal implant device. Such a system may be used to evaluate and optimize competing algorithms.

Methods: : The system is based on the Video Surveillance Kit available from AVNET Electronics. This kit includes a CMOS sensor capable of capturing 752x480 pixel images, a Digital Signal Processor, and USB and network ports used to communicate with the board. Images are captured by the CMOS sensor and analyzed by the on board DSP to extract features such as image boundaries, motion, average intensity, etc. Such features are transmitted to a desktop PC for offline analysis.

Results: : The CMOS sensor plus DSP combination are capable of capturing images and extracting features in real time. The hardware and software provide a framework for testing and evaluating a variety of feature extraction algorithms.

Conclusions: : This system is sufficient for developing, evaluating, and optimizing a variety of feature extraction algorithms. It will need to be miniaturized and made to operate in a standalone fashion (i.e. without the desktop PC) in order to be incorporated into a functional portable prosthetic vision system, but such a task is a fairly straightforward engineering effort.

Keywords: retina • scene perception • quality of life